Editorial Team
8 min read
Nebius Group revenue rose 684% to $399 million in the first quarter of 2026 while purchases of property, equipment and intangible assets reached $2.47 billion. Microsoft and Meta reduce demand risk, but options and an unsold-capacity backstop leave delivery, financing and unit economics unresolved.
Nebius has evidence of demand that many infrastructure developers would envy. What it does not yet have is public evidence that the capacity behind its largest agreements can be delivered on schedule and earn attractive returns after construction, hardware, depreciation and financing costs.
Magica visualization of company-reported consolidated revenue and total operating costs and expenses for the quarters ended March 31, 2025 and 2026. Results include Nebius, Avride and TripleTen. Source: U.S. Securities and Exchange Commission.
Nebius Group’s unaudited first-quarter release reported $399 million of revenue for the three months ended March 31, 2026, up from $50.9 million a year earlier. Adjusted EBITDA, a company-defined non-GAAP measure, improved to $129.5 million from a $53.7 million loss.
The GAAP results give a more guarded view of operating performance. Nebius recorded a $128 million operating loss and a $100.3 million adjusted net loss. Its $621.2 million GAAP net income was more than explained by a $780.6 million gain from revaluing equity investments, not by a profit from operations.
At the same time, purchases of property, equipment and intangible assets reached $2.473 billion, or 6.2 times quarterly revenue. That ratio is a snapshot of a construction quarter, not a forecast of steady-state economics, but it shows the amount of infrastructure that must be financed before it can generate service revenue. Depreciation and amortization was already $212 million, equal to 53% of revenue.
Customer funding and external capital helped carry the outlay. Cash provided by continuing operations was $2.258 billion as deferred revenue contributed $3.198 billion to operating cash flow. Financing activities supplied another $6.296 billion, including $4.338 billion of convertible-note proceeds before issuance costs and $2 billion from prefunded warrants.
Nebius therefore finished March with $9.298 billion in cash, but also with $8.450 billion of current and non-current debt, up from $4.128 billion at the end of 2025. Current and non-current deferred revenue totaled $4.778 billion. Those balances demonstrate access to liquidity and customer commitments; they do not show that the operating business can yet finance expansion from retained earnings.
The figures are also group results. They consolidate the core AI cloud business with autonomous-vehicle developer Avride and education service TripleTen, so they are not a clean margin statement for the cloud operation alone.
Nebius’s two largest disclosed customer arrangements combine firm orders, optional growth and a demand backstop. Treating their maximum value as a single committed-revenue figure erases those differences.
The Microsoft filing describes a five-year GPU-services agreement with a value of about $17.4 billion through 2031, subject to deployment and availability. Microsoft may acquire more services or capacity, which would lift the total to about $19.4 billion. The filing says cash flow from the agreement will finance part of the associated capital spending.
The Meta filing separates $12 billion of orders from a second arrangement worth up to $15 billion. Nebius intends to offer specified clusters to third-party cloud customers first. If some of that capacity remains unsold, Meta must buy it for the remainder of a period ending five years after deployment.
| Component | Disclosed value | Economic character |
|---|---|---|
| Microsoft base agreement | About $17.4 billion | GPU services, subject to deployment and availability |
| Microsoft expansion | Up to about $2 billion | Additional services or capacity Microsoft may acquire |
| Meta orders | $12 billion | Capacity plus associated storage and connectivity |
| Meta backstop | Up to $15 billion | Meta purchases specified capacity left unsold to other customers |
The two base amounts total about $29.4 billion. The maximum reaches about $46.4 billion only if Microsoft takes the additional services or capacity and the Meta backstop is used in full. Third-party sales and Meta backstop purchases for the same clusters are alternatives, not additive revenue.
An independent account also identifies an earlier $3 billion Meta agreement signed in November 2025. The retained materials do not establish how every analyst’s backlog figure treats that earlier agreement, so mechanically adding it to the $46.4 billion ceiling would create false precision.
The contracts lower demand risk but leave deployment risk with Nebius. Microsoft can terminate an undelivered GPU service after a grace period if Nebius cannot provide alternative capacity, and the agreement includes liquidated damages for delay. The parties’ obligations start only when Nebius confirms that it has secured any additional financing required for the capital expenditure. The Meta arrangements include discounted monthly fees for late delivery.
Neither filing discloses a unit price, expected margin, return on invested capital or the cost to deliver each cluster. The Meta backstop establishes a buyer for otherwise unsold capacity, but the public terms do not show whether selling to Meta or to third parties produces better economics.
Founder and CEO Arkady Volozh said on an earnings call that everything Nebius builds is sold, according to a syndicated market analysis. The same account said the core AI business generated $390 million of first-quarter revenue, up 841% year over year, and that Nebius raised the midpoint of its capital-spending guidance from $18 billion to $22.5 billion.
That demand sits within a market in which large technology companies are supplementing their own data centers with scarce GPU and power capacity from specialists. CoreWeave is a direct alternative named in the independent account, while large cloud providers already serve a broader customer base. The headline of a retained item says Wolfe Research sees CoreWeave as “the best stock to own” in a note concerning neoclouds. Its archived body supplies no rationale, rating details or financial comparison, so it cannot support more specific conclusions about Wolfe’s reasoning.
The company has another source of strategic capital: independent reporting says Nvidia announced in March that it would invest $2 billion for an 8.3% stake in Nebius. But capital access does not remove infrastructure constraints. In its financial release, Nebius listed timely and cost-effective data-center development, access to power, equipment supply, dependence on key vendors, technological change and additional financing among the risks to its plans. It also announced that it had secured up to 1.2 gigawatts of power and land for a planned Pennsylvania site; “up to” capacity at a planned facility is not the same as commissioned computing capacity.
Nebius itself is a recently recast company. The syndicated account traces it to Yandex and says the Amsterdam-based holding company relaunched as Nebius in 2024 after selling Yandex’s Russian operations for $5.3 billion. That gives Nebius a large-technology lineage, but the retained materials show only a short record under its current cloud identity.
The most expansive projection in the retained material comes from the syndicated market analysis, not from Nebius’s filings or the available Wolfe Research item. It says analysts tracking the stock expect revenue to rise from $3.36 billion in 2026 to $34 billion in 2030, with more than $115 billion of capital expenditure over the period and positive free cash flow of $1.37 billion only in 2030.
Moving from $3.36 billion to $34 billion in four years requires compound annual growth of about 78%. The calculation is straightforward; the forecast’s construction is not. The article does not identify which analysts produced the estimates, whether the figures are a consensus, how they treat the contract options and Meta backstop, or what pricing, utilization, depreciation and financing assumptions they use.
The comparison base also matters. The $3.36 billion figure is a full-year revenue estimate. Nebius separately said in the independent account that it expected an annualized revenue run rate of $7 billion to $9 billion at the end of 2026. A year-end run rate is a point-in-time pace, not revenue earned across the year, so it cannot be compared directly with either $3.36 billion or $34 billion without a ramp schedule.
The $115 billion capital-spending projection is similarly an estimate, not an obligation disclosed in the customer filings. It may convey the scale required by the revenue forecast, but without annual phasing, capacity assumptions or a return threshold, it cannot establish whether the forecast creates value.
The next useful disclosure is not another maximum contract value. It is evidence connecting construction to service revenue and service revenue to cash returns. Four measurements would narrow the uncertainty:
The current record supports a narrower conclusion than either a simple backlog celebration or a claim that demand is the main problem. Nebius has customers, cash and access to financing, while Meta has absorbed part of the unsold-capacity risk. The unresolved issue is whether Nebius can turn those advantages into commissioned infrastructure and positive cash economics after accounting for delivery remedies, depreciation and financing costs. Until those results are disclosed, $34 billion remains an undocumented analyst endpoint rather than evidence of an achieved trajectory.
Kimi says a 48-hour demand surge pushed its GPU capacity close to the limit, forcing a pause in new subscriptions as a separate Kimi Code plan prepares to change how coding access is bundled and rationed.
Ant Digital Technologies has expanded Agentar with 200 preconfigured job-role templates and a multi-agent management pitch, but it has not disclosed pricing, customer use or performance data—and governance is becoming a market-wide requirement rather than a distinctive feature.
Apple is reportedly testing an opt-in tool that transcribes and summarizes Genius Bar appointments. Its current safeguards are clear, but its accuracy, retention rules and uses after the pilot are not.
Moonshot AI is seeking investor approval for a Hong Kong IPO within six months while an unfinished private round could value it above $30 billion. The pitch pairs rapid reported recurring-revenue growth with Kimi K3, but neither audited financials nor enough independent model and deployment evidence is public yet.
Andy Serkis says machine learning has a narrow role in The Hunt for Gollum’s de-aging work. That boundary remains a production claim, because the actors, shots, tools, data, labor effects and likeness terms have not been disclosed.
Chinese-developed models have overtaken U.S. rivals in token volume on OpenRouter, where low prices and token-heavy agent workloads favor DeepSeek V4. The crossover covers a small, platform-specific slice of AI use and does not establish leadership in revenue, enterprise demand or infrastructure control.
OpenAI said it had identified the cause of elevated ChatGPT errors and was applying mitigations, but the retained status update did not disclose the technical failure, measure the impact or confirm full recovery.
Alibaba has opened hosted access to Qwen3.8-Max-Preview and says the 2.4-trillion-parameter model will be released with downloadable weights, but the license, architecture, benchmark evidence, release date and model-level credit economics remain undisclosed.
Anthropic has put Bun’s Rust port into Claude Code ahead of Bun 1.4’s general release, creating a real but tightly controlled proving ground for an AI-led migration whose total cost and broader reliability remain unsettled.
Zhipu reportedly reached $1 billion in annual recurring revenue in July, roughly four times a March estimate, but the unconfirmed run rate is not annual sales and still sits far ahead of recognized cloud revenue while margins remain thin.
An account of PNC transaction data puts household-paid generative AI near 2%, while a separate user survey finds much broader paid access when employer-funded plans count. The gap shows why card charges alone cannot settle whether consumer AI is becoming a mass subscription business.
Kimi K3 reduces attention traffic, but Moonshot recommends deploying it across at least 64 accelerators. SemiAnalysis says expert routing will more than erase the bandwidth savings; until the promised weights are deployed independently, that remains a hardware thesis rather than a measured result.
Morgan Stanley raised its Micron fiscal-2027 gross-margin estimate to 89.3%, but Micron’s results show the forecast depends chiefly on exceptional memory pricing, customer contracts and delayed supply rather than a disclosed HBM4 margin advantage.
New Mexico’s land commissioner refused to reconsider state-land crossings for a pipeline serving Project Jupiter, preserving a fuel-supply obstacle for the planned Oracle data center. But an analyst’s 2029 forecast predates that decision and remains at odds with Oracle’s first-half-2027 delivery statement.
A reported CIA mission examined whether an influential Emirati sheikh could be trusted with sensitive U.S. technology. The public export rule that followed gives G42 and Core42 a narrow, temporary exception, but it does not connect the intelligence operation to that decision or show how compliance will be tested.
Tracebit says a guardrail-triggering string in one decoy AWS secret sharply reduced five AI agents' success in a 152-run cyber range, but the company-run test did not cover uncensored models, adaptive attackers or production deployments.
SenseTime’s U1 Pro preview combines a claimed native 8K ceiling with a multi-step image-creation loop, but its August API will need to disclose dimensions, latency, pricing and repeatable results before buyers can compare the cost of a usable asset.
Alibaba Cloud has begun invite-only testing of a 64-card Zhenwu M890 supernode instance, extending its in-house chip and infrastructure stack to outside users while leaving the service's price, benchmark methodology and customer economics undisclosed.
Nvidia's 616.00 driver and CUDA 13.4 preview let developers begin native Windows Arm64 work for RTX Spark, but the release is an ecosystem-building step—not evidence of final performance, compatibility or pricing.
Open Design has attracted nearly 80,000 GitHub stars with an open, model-flexible answer to Claude Design, but its million-install claim has no published methodology and the team has not disclosed the retention and revenue figures needed to judge the business.